The IP over ATM Mailing List Archive by date

[manfredi@engr05.comsys.rockwell.com (Albert E. Manfredi)]

> From:	J.Crowcroft@cs.ucl.ac.uk    29-MAR-1996 02:57
 
> if (a big if) you allocate a separate VCI/VPI for each source in a multipoint
> session, then it can index the analagous state as the source does in IP
> multicast quite happily.....if this state indicates a _set_ of sources
> associated with a multipoint destination (which is the case, since
> there is the conversation id used in the call setup), then you could
> do the resequencing at the last switch before each receiver....or in
> the receiver h/w....

Exactly! If this is really a mpt-mpt conversation, then that one receiver 
that does the one-time-only resequencing might be an MCS.

> or, you could allocate a VPI for the group, and a VCI for each source
> if only there wasn't such a small VCI/VPI address space...

I just don't see this as a problem. The actual _value_ of the VPI/VCI 
doesn't have to be the same for all participants of the multicast. They 
would be the same but possibly labeled differently, which is the way ATM 
was always supposed to handle VPI/VCI. Given that, I doubt you can make a 
case that the number of possibilities is too limited. You're going to run 
out of backplane before anything else.

> actually, itd be interesting to know how many switches actually have a
> large cell laberl space internally, since its irrelvant what is seen
> at the UNI for a circuit based net - the associateion between the
> label and the actual multipoint group is done at call time, so so long
> as we could increase the address space in the NNI, we could do the
> multipoint-multipoint routing quite neatly - so here is a proposal
> 
> instead of stealing precious data space from the user by proposing a
> MID for AAL5 (AAL5++, or AAL5 double cross as i heard someone refer to
> it:-), we propose an _encapsulation_ of the cell in a multipoint
> distribution protocol with a new global
> group address header stuck on the front....(and the VCI/VPI address
> space is then used for source identification - or vice versa - no
> special preference...)

A bigger cell? Global group address carried _with_ each cell? Heresy! Are 
you trying to do a sex change unto ATM or something?

> of course there are those that argue that there is no need for so many
> active sources, and so they can use the VPI/VCI hack above
> instead...or suffer the signaling traffic...

My vote is with the hack.

Bert
manfredi@engr05.comsys.rockwell.com